Date: Tue, 02 Jun 1998 14:29:26 +0000 (GMT)
From: Andrew Brown <a.brown-AT-mdx.ac.uk>
Subject: BHA: defining open / closed systems: implications for research


Dear Doug and list members,

First I continue to 'lurk and learn' a great deal from this list - 
thanks to all.

Second, the title of Doug's paper prompted me to brush up an 
aborted reply to an exchange we had some months ago on the list. Of 
general interest is Doug's suggestion that closed systems be defined 
as those containing invariant empirical regularities. I argue below 
that this definition does not square with RB's work and suggest that 
structural interaction must be the explicit key to defining the 
closed / open distinction. I go on to reconsider the issue of simple 
and multiple regression on the basis established. I consider 
Jessop's regulation theory and US SSA theory along the way. Doug, I'm 
very greatful for your earlier replies which have lead to the rethink 
below. 

On the key topic of the definition and nature of open systems, Doug 
argued that empirical regularities are not rare in the social or 
natural world. What ARE rare are *invariant* empirical regularities 
on Doug's interpretation of RB. Therefore Doug suggested that the 
presence of invarient empirical regularities defines closed systems 
for RB.

Now, as a critique of the covering law model then the definition may
be very useful and apt. If that model does imply invariant
empirical regularities then it must be wrong. However at least two 
arguments count against the definition. I should emphasise that 
with each argument I am primarily concerned with interpreting and 
assessing the practical implications of RB's work.

(1) The definition has no methodological import; 
specifically it would make the closed/open system distinction 
irrelevant for the issue of the need for, and nature of, experiment. 
This is because 'invariant regularity' is synonymous with 'eternal 
regularity' and a regularity that lasts for a million years is no 
more eternal than one which lasts for two seconds. In the case of 
'open systems' with long lasting empirical regularities science would 
have absolutely no need for the creation of closed systems by 
experiment. Indeed the very idea would be rather absurd. Yet as far 
as RB is concerned the open / closed distinction IS crucial to 
understand the nature of, and need for, experiment (RTS). This is 
due, of course, to the actual interaction of structures (I will spell 
this out below). It is the lack of experiment in social science leads 
RB to search, in PON ch. 2, for a method that *does not involve 
empirical regularities at all* (the method he suggests rests on the 
notion of preconceptualised social forms and activities).

(2) According to RB *Social* structures and mechanisms are NOT 
eternal, they are only 'relatively enduring' so social science can 
never find eternal regularities.

Note that it is the fact that structures interact that prevents (for 
the most part) empirical regularities outside of experiment; the 
length of endurance of structures is irrelevant to this issue. Indeed 
the shorter life of social structures would lead one to expect that 
the social world contains less regularities than the natural one. 
[Thus the attempted justification of the use of stylised facts / 
demi-regularities on the grounds that social structures and 
mechanisms are only relatively enduring, producing only a 'partial' 
regularity, is invalid. Relative endurance has nothing to do with it. 
There should only rarely be any significant regularities in the 
social world.] 

The above would seem to suggest that RB could make the interaction
of different mechanisms a defining feature of open systems rather 
than the absence of invariant regularities.

The arguments need not *rule out* the use of simple 
regression completely. If 'demi-regularities' should occur then it 
may be valid to use them for retroduction. However any social theory 
resting heavily upon (even together with a range of methods) 
'stylised facts' or 'demi-regularities' for retroduction must be 
given a special justification which can only be a specification of 
the rare ontological feature leading to an empirical regularity. 

A particular example that I would cite as transgressing the limits 
imposed by RB's arguments are the regulation theorists, at least as 
interpreted by Jessop and the human geographers. Regulation 
theory is constructed, according to Jessop, by retroduction from 
stylised facts (demi-regularities?) such as the purported high wage / 
high productivity figures of the 1950's and 1960's. There is no sense 
of ontological peculiarity / rarity here.

As for multiple regression it seems that RB's work would provide 
support for it, subject to one crucial distinction which I failed to 
make explicit in earlier posts. This distinction is that between the 
retroduction of underlying social structures on the one hand and the 
estimation of structural parameters on the other. The key difficulty 
RB would point to for social science is the construction of concepts 
of underlying structures and this cannot be much helped by 
multiple regression due to the actual interaction of structures. But 
once adequate concepts of structures have been *constructed* their 
relevant quantitative aspects can be entered as explanatory variables 
in multiple regression analysis with actual variables as explained 
variables.

So a RB multiple regression could not uncritically enter actual 
variables as explanatory variables rather it must *construct* 
explanatory variables as the quantitative aspects of retroduced 
structures. 

Incidentally, the American 'social structures of accumulation' 
theorists (Bowles, Gintis, Weiskopf, Gordon etc.) claim precisely to 
*construct* an 'unusual data set' which they claim captures 
underlying structural / institutional explanatory variables. So RB 
can have no methodological quibble with their use of multiple 
regression. The RB grounds for criticism of the SSA regressions might 
rest on the procedure by which SSA theorists construct their 
conception of underlying structures (e.g. they have been accused of 
methodological individualism). Otherwise there would be no 
philosophical or methodological grounds to criticise SSA work. SSA's 
relative merits would rest on entirely substantive issues for RB. 
Similar considerations would apply to orthodox econometrics where it 
would be methodological individualism and actualism rather than the 
use multiple regression per se that breaks RB precepts.

Many thanks,
andy.


     --- from list bhaskar-AT-lists.village.virginia.edu ---